TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC
COHOMOLOGY
NILS A. BAAS, BJØRN IAN DUNDAS AND JOHN ROGNES
Dedicated to Graeme Segal on the occasion of his 60th birthday
1. Introduction
The work to be presented in this paper has been inspired by several of
Professor Graeme Segal's papers. Our search for a geometrically defined
elliptic cohomology theory with associated elliptic objects obviously stems
from his Bourbaki seminar [28]. Our readiness to form group completions
of symmetric monoidal categories by passage to algebraic K-theory spectra
derives from his Topology paper [27]. Our inclination to invoke 2-functors
to the 2-category of 2-vector spaces generalizes his model for topological K-
theory in terms of functors from a path category to the category of vector
spaces. We offer him our admiration.
Among all generalized (co-)homology theories, a few hold a special po-
sition because they are, in some sense, geometrically defined. For exam-
ple, de Rham cohomology of manifolds is defined in terms of cohomology
classes of closed differential forms, topological K-theory of finite CW com-
plexes is defined in terms of equivalence classes of complex vector bundles,
and complex bordism is defined in terms of bordism classes of maps from
stably complex manifolds. The geometric origin of these theories makes
them particularly well suited to the analysis of many key problems. For
example, Chern-Weil theory associates differential forms related to the cur-
vature tensor to manifolds with a connection, whose de Rham cohomology
classes are the Chern classes of the tangent bundle of the manifold. The
Atiyah-Segal index theory [2] associates formal differences of vector bundles
to parametrized families of Fredholm operators, arising e.g. from complexes
of elliptic pseudo-differential operators, and their isomorphism classes live
in topological K-theory. Moduli spaces of isomorphism classes of solutions
to e.g. Yang-Mills gauge-theoretic problems can generically yield maps from
suitably structured manifolds, with well-defined bordism classes in the corre-
sponding form of bordism homology. On the other hand, Quillen's theorem
that the coefficient ring ß*(MU) for complex bordism theory is the Lazard
ring that corepresents (commutative 1-dimensional) formal group laws has
no direct manifold-geometric interpretation, and may seem to be a fortuitous
coincidence in this context.
1
2 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
From the chromatic point of view of stable homotopy theory, related to
the various periodicity operators vn for n 0 that act in many cohomo-
logy theories, these three geometrically defined cohomology theories detect
an increasing amount of information. De Rham cohomology or real singular
cohomology sees only rational phenomena, because for each prime p multi-
plication by p = v0 acts invertibly on H*(X; R). Topological K-theory only
picks up Bott periodic phenomena, because multiplication by the Bott class
u 2 ß2(KU) acts invertibly on KU*(X), and up-1 = v1 for each prime p.
Complex bordism MU*(X) instead detects all levels of periodic phenomena.
We can say that real cohomology, topological K-theory and complex bordism
have chromatic filtration 0, 1 and 1, respectively. A precise interpretation
of this is that the spectra HR and KU are Bousfield local with respect to
the Johnson-Wilson spectra E(n) for n = 0 and 1, respectively, while MU
is not E(n)-local for any finite n. Traditionally, an elliptic cohomology the-
ory is a complex oriented Landweber exact cohomology theory associated to
the formal group law of an elliptic curve. It will have chromatic filtration 2
when the elliptic curve admits a supersingular specialization, and so any
cohomology theory of chromatic filtration 2 might loosely be called a form
of elliptic cohomology. However, the formal group law origin of traditional
elliptic cohomology is not of a directly geometric nature, and so there has
been some lasting interest in finding a truly geometrically defined form of
elliptic cohomology.
It is the aim of the present paper to introduce a geometrically defined
cohomology theory that is essentially of chromatic filtration 2, or more pre-
cisely, a connective form of such a theory. It therefore extends the above
list of distinguished cohomology theories one step beyond topological K-
theory, to a theory that will detect v2-periodic phenomena, but will ignore
the complexity of all higher vn-periodicities for n 3.
The theory that we will present is represented by the algebraic K-theory
spectrum K(V) of the Kapranov-Voevodsky 2-category of 2-vector spaces
[15]. A 2-vector space is much like a complex vector space, but with all
occurrences of complex numbers, sums, products and equalities replaced by
finite-dimensional complex vector spaces, direct sums, tensor products and
coherent isomorphisms, respectively. It is geometrically defined in the sense
that the 0-th cohomology group K(V)0(X) of a space X can be defined in
terms of equivalence classes of 2-vector bundles over X (or more precisely,
over the total space Y of a Serre fibration Y ! X with acyclic homotopy
fibers, i.e., an acyclic fibration). Cf. theorem 4.10. A 2-vector bundle over X
is a suitable bundle of categories, defined much like a complex vector bundle
over X, but subject to the same replacements as above. The previously
studied notion of a gerbe over X with band C* is a special case of a 2-vector
bundle, corresponding in the same way to a complex line bundle.
We conjecture in 5.1 that the spectrum K(V) is equivalent to the al-
gebraic K-theory spectrum K(ku) of the connective topological K-theory
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 3
spectrum ku, considered as a "brave new ring", i.e., as an S-algebra. This
is a special case of a more general conjecture, where for a symmetric bi-
monoidal category B (which is a generalization of a commutative semi-ring)
we compare the category of finitely generated free modules over B to the
category of finitely generated free modules over the commutative S-algebra
A = Spt(B) (which is a generalization of a commutative ring) associated
to B. The conjecture amounts to a form of öp sitive thinking", asserting
that for the purpose of forming algebraic K-theory spectra it should not
matter whether we start with a semi-ring-like object (such as the symmetric
bimonoidal category B) or the ring-like object given by its additive Groth-
endieck group completion (such as the commutative S-algebra A). This idea
originated with Marcel Bökstedt, and we are indebted to him for suggesting
this approach. We have verified the conjecture in the case of actual commu-
tative semi-rings, interpreted as symmetric bimonoidal categories that only
have identity morphisms, and view this as strong support in favor of the
conjecture.
Continuing, we know that K(ku), or rather a spectrum very closely re-
lated to it, is essentially of chromatic filtration 2. For connective spectra,
such as all those arising from algebraic K-theory, there is a more appropri-
ate and flexible variation of the chromatic filtration that we call the tele-
scopic complexity of the spectrum; cf. definition 6.1. For example, integral
and real cohomology have telescopic complexity 0, connective and periodic
topological K-theory have telescopic complexity 1, and traditional elliptic
cohomology has telescopic complexity 2.
It is known, by direct nontrivial calculations [3], that K(`^p) has telescopic
complexity 2, where `^pis the connective p-complete Adams summand of
topological K-theory and p 5. The use of the Adams summand in place
of the full connective p-complete topological K-theory spectrum ku^p, as
well as the hypothesis p 5, are mostly technical assumptions that make
the calculations manageable, and it seems very likely that also K(ku^p) will
have telescopic complexity 2 for any prime p. It then follows from [9], if we
assume the highly respectable Lichtenbaum-Quillen conjecture for K(Z) at
p, that also K(ku) has telescopic complexity 2. In this sense we shall allow
ourselves to think of K(ku), and conjecturally K(V), as a connective form
of elliptic cohomology.
The definition of a 2-vector bundle is sufficiently explicit that it may carry
independent interest. In particular, it may admit notions of connective struc-
ture and curving, generalizing the notions for gerbes [8, x5.3], such that to
each 2-vector bundle E over X with connective structure there is an asso-
ciated virtual vector bundle H over the free loop space LX = Map (S1, X),
generalizing the anomaly line bundle for gerbes [8, x6.2]. If E is equipped
with a curving, there probably arises an action functional for oriented com-
pact surfaces over X (loc. cit.), providing a construction of an elliptic object
over X in the sense of Segal [28]. Thus 2-vector bundles over X (with extra
structure) may have naturally associated elliptic objects over X. However,
4 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
we have not yet developed this theory properly, and shall therefore postpone
its discussion to a later joint paper, which will also contain proofs of the re-
sults announced in the present paper. Some of the basic ideas presented
here were sketched by the first author in [4].
The paper is organized as follows. In x2 we define a charted 2-vector bun-
dle of rank n over a space X with respect to an open cover U that is indexed
by a suitably partially ordered set I. This corresponds to a Steenrod-style
definition of a fiber bundle, with standard fiber the category Vn of n-tuples
of finite-dimensional complex vector spaces, chosen trivializations over the
chart domains in U, gluing data that compare the trivializations over the
intersection of two chart domains and coherence isomorphisms that system-
atically relate the two possible comparisons that result over the intersection
of three chart domains. We also discuss when two such charted 2-vector bun-
dles are to be viewed as equivalent, i.e., when they define the same abstract
object.
In x3 we think of a symmetric bimonoidal category B as a generalized
semi-ring, and make sense of the algebraic K-theory K(B) of its 2-category
of finitely generated free öm dules" Bn. We define the weak equivalences
Bn ! Bn to be given by a monoidal category M = GLn(B) of weakly
invertible matrices over B, cf. definition 3.6, in line with analogous con-
structions for simplicial rings and S-algebras [33]. It is a key point that we
allow GLn(B) to contain more matrices than the strictly invertible ones, of
which there are too few to yield an interesting theory. We also present an
explicit bar construction BM that is appropriate for such monoidal cate-
gories. Our principal example is the symmetric bimonoidal category V of
finite-dimensional complex vector spaces under direct sum and tensor prod-
uct, for which the modules Vn are the 2-vector spaces of Kapranov and
Voevodsky.
In x4 we bring these two developments together, by showing that the
equivalence classes of charted 2-vector bundles of rank n over a (reasonable)
space X is in natural bijection (theorem 4.5) with the homotopy classes of
maps from X to the geometric realization |BGLn(V)| of the bar construction
on the monoidal category of weakly invertible n x n matrices over V. The
group of homotopy classes of maps from X to the algebraic K-theory space
K(V) is naturally isomorphic (theorem 4.10) to the Grothendieck group
completion of the abelian monoid of virtual 2-vector bundles over X, i.e.,
the 2-vector bundles E # Y over spaces Y that come equipped with an acyclic
fibration a: Y ! X. Hence the contravariant homotopy functor represented
by K(V) is geometrically defined, in the sense that virtual 2-vector bundles
over X are the (effective) cycles for this functor at X.
In x5 we compare the algebraic K-theory of the generalized semi-ring
B to the algebraic K-theory of its additive group completion. To make
sense of the latter as a ring object, as is necessary to form its algebraic K-
theory, we pass to structured ring spectra, i.e., to the commutative S-algebra
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 5
A = Spt(B). We propose that the resulting algebraic K-theory spectra
K(B) and K(A) are weakly equivalent (conjecture 5.1), and support this
assertion by confirming that it holds true in the special case of a discrete
symmetric bimonoidal category B, i.e., a commutative semi-ring in the usual
sense. In the special case of 2-vector spaces the conjecture asserts that
K(V) is the algebraic K-theory K(ku) of connective topological K-theory
ku viewed as a commutative S-algebra.
In x6 we relate the spectrum K(ku) to the algebraic K-theory spectrum
K(`^p) of the connective p-complete Adams summand `^pof ku^p. The latter
theory K(`^p) is known (theorem 6.4, [3]) to have telescopic complexity 2, and
this section makes it plausible that also the former theory K(ku) has tele-
scopic complexity 2, and hence is a connective form of elliptic cohomology.
Together with conjecture 5.1 this says that (a) the generalized cohomology
theory represented by K(ku) is geometrically defined, because its 0-th co-
homology group, which is then represented by K(V), is defined in terms
of formal differences of virtual 2-vector bundles, and (b) that it has tele-
scopic complexity 2, meaning that it captures one more layer of chromatic
complexity than topological K-theory does.
2. Charted two-vector bundles
Definition 2.1. Let X be a topological space. An ordered open cover (U, I)
of X is a collection U = {Uff| ff 2 I} of open subsets Uff X, indexed by a
partially ordered set I, such that
S
(1) the Uffcover X in the sense that ffUff= X, and
(2) the partial ordering on I restricts to a total ordering on each finite
subset {ff0, . .,.ffp} of I for which the intersection Uff0...ffp= Uff0\
. .\.Uffpis nonempty.
The partial ordering on I makes the nerve of the open cover U an ordered
simplicial complex, rather than just a simplicial complex. We say that U is a
good cover if each finite intersection Uff0...ffpis either empty or contractibl*
*e.
Definition 2.2. Let X be a topological space, with an ordered open cover
(U, I), and let n 2 N = {0, 1, 2, . .}.be a non-negative integer. A charted
2-vector bundle E of rank n over X consists of
(1) an n x n matrix
Efffi= (Efffiij)ni,j=1
of complex vector bundles over Ufffi, for each pair ff < fi in I, such
that over each point x 2 Ufffithe integer matrix of fiber dimensions
dim(Efffix) = (dim Efffiij,x)ni,j=1
is invertible, i.e., has determinant 1, and
(2) an n x n matrix
~=
OEfffifl= (OEfffiflik)ni,k=1:Efffi.-Efifl---!Efffl
6 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
of vector bundle isomorphisms
L n fffi fifl~= fffl
OEfffiflik:j=1Eij Ejk----! Eik
over Ufffifl, for each triple ff < fi < fl in I, such that
(3) the diagram
ff_
Efffi. (Efifl. Eflffi)_________//(Efffi. Efifl) . Eflffi
id.ffififlffi|| |ffifffifl.id|
fflffl| fflffl|
Efffi. Efiffiffifffiffi//_EffffiEfffl.fEflffififfflffioo_
of vector bundle isomorphisms over Ufffiflfficommutes, for each chain
ff < fi < fl < ffi in I.
Here ff_denotes the (coherent) natural associativity isomorphism for the
matrix product . derived from the tensor product of vector bundles. We
call the n x n matrices Efffiand OEfffiflthe gluing bundles and the coherence
isomorphisms of the charted 2-vector bundle E # X, respectively.
Definition 2.3. Let E and F be two charted 2-vector bundles of rank n
over X, with respect to the same ordered open cover (U, I), with gluing bun-
dles Efffiand F fffiand coherence isomorphisms OEfffifland _fffifl, respectivel*
*y.
An elementary change of trivializations (T ff, øfffi) from E to F is given by
(1) an n x n matrix T ff= (Tiffj)ni,j=1of complex vector bundles over Uff,
for each ff in I, such that over each point x 2 Uffthe integer matrix
of fiber dimensions dim(Txff) has determinant 1, and
(2) an n x n matrix of vector bundle isomorphisms
~=
øfffi= (øfffiij)ni,j=1:F fffi.-T-fi--!T ff. Efffi
over Ufffi, for each pair ff < fi in I, such that
(3) the diagram
id.fififl//_ fifffi.id//_
F fffi. F fifl. T flF fffi. T fi. EfiflT ff. Efffi. Efifl
_fffifl.id|| id.ffifffifl||
fflffl| fflffl|
F fffl. T_fl_________fifffl__________//T ff. Efffl
(natural associativity isomorphisms suppressed) of vector bundle iso-
morphisms over Ufffiflcommutes, for each triple ff < fi < fl in I.
Definition 2.4. Let (U, I) and (U0, I0) be two ordered open covers of X.
Suppose that there is an order-preserving carrier function c: I0 ! I such
that for each ff 2 I0 there is an inclusion U0ff Uc(ff). Then (U0, I0) is a
refinement of (U, I).
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 7
Let E be a charted 2-vector bundle of rank n over X with respect to (U, I),
with gluing bundles Efffiand coherence isomorphisms OEfffifl. Let
c*Efffi= Ec(ff)c(fi)|U0fffi
for ff < fi in I0and
c*OEfffifl= OEc(ff)c(fi)c(fl)|U0fffifl
for ff < fi < fl in I0, be n x n matrices of vector bundles and vector bundle
isomorphisms over U0fffiand U0fffifl, respectively. Then there is a charted 2-
vector bundle c*E of rank n over X with respect to (U0, I0), with gluing
bundles c*Efffiand coherence isomorphisms c*OEfffifl. We say that c*E is an
elementary refinement of E.
More generally, two charted 2-vector bundles of rank n over X are said to
be equivalent 2-vector bundles if they can be linked by a finite chain of ele-
mentary changes of trivializations and elementary refinements. (This is the
notion of equivalence that appears to be appropriate for our representability
theorem 4.5.)
Remark 2.5. A charted 2-vector bundle of rank 1 consists of precisely the
data defining a gerbe over X with band C*, as considered e.g. by Giraud [10],
Brylinski [8] and Hitchin [13, x1]. There is a unitary form of the definition
above, with Hermitian gluing bundles and unitary coherence isomorphisms,
and a unitary 2-vector bundle of rank 1 is nothing but a gerbe with band
U(1). In either case, the set of equivalence classes of C*-gerbes or U(1)-
gerbes over X is in natural bijection with the third integral cohomology
group H3(X; Z) [8, 5.2.10].
Definition 2.6. Let E # X be a charted 2-vector bundle of rank n, with
notation as above, and let a: Y ! X be a map of topological spaces. Then
there is a charted 2-vector bundle a*E # Y of rank n obtained from E by
pullback along a. It is charted with respect to the ordered open cover (U0, I)
with U0 = {U0ff= a-1(Uff) | ff 2 I}. It has gluing bundles a*Efffiobtained
by pullback of the matrix of vector bundles Efffialong a: U0fffi! Ufffi, and
coherence isomorphisms a*OEfffiflobtained by pullback of the matrix of vector
bundle isomorphisms OEfffiflalong a: U0fffifl! Ufffifl. By definition there is
then a map of charted 2-vector bundles ^a:a*E ! E covering a: Y ! X.
Definition 2.7. Let E # X and F # X be charted 2-vector bundles with
respect to the same ordered open cover (U, I) of X, of ranks n and m, with
gluing bundles Efffiand F fffiand coherence isomorphisms OEfffifland _fffifl,
respectively. Their Whitney sum E F # X is then the charted 2-vector
bundle of rank (n + m) with gluing bundles given by the (n + m) x (n + m)
matrix of vector bundles ` '
Efffi 0
0 F fffi
8 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
and coherence isomorphisms given by the (n+m)x(n+m) matrix of vector
bundle isomorphisms
` ' ` ' ` ' ` '
OEfffifl0 : Efffi 0 . Efifl 0 --~=--! Efffl 0 .
0 _fffifl 0 F fffi 0 F fifl 0 F fffl
There is an elementary change of trivializations from E F to F E given
by the (n + m) x (n + m) matrix
` '
T ff= I0 Im
n 0
for each ff in I, and identity isomorphisms øfffi. Here In denotes the identity
n x n matrix, with the trivial rank 1 vector bundle in each diagonal entry
and zero bundles elsewhere.
3. Algebraic K-theory of two-vector spaces
Let (B, , , 0_, 1_) be a symmetric bimonoidal category, with sum and
tensor functors
, : B x B ----! B ,
and zero and unit objects 0_, 1_in B. These satisfy associative, commutative
and distributive laws, etc., up to a list of natural isomorphisms, and these
isomorphisms are coherent in the sense that they fulfill a (long) list of com-
patibility conditions, as presented by Laplaza in [17, x1]. We say that B
is a bipermutative category if the natural isomorphisms are almost all iden-
tity morphisms, except for the commutative laws for and and the left
distributive law, and these in turn fulfill the (shorter) list of compatibility
conditions listed by May in [20, xVI.3].
Suppose that B is small, i.e., that the class of objects of B is in fact a
set. Let ß0(B) be the set of path components of the geometric realization
of B. (Two objects of B are in the same path component if and only if
they can be linked by a finite chain of morphisms in B.) Then the sum
and tensor functors induce sum and product pairings that make ß0(B) into
a commutative semi-ring with zero and unit. We can therefore think of
the symmetric bimonoidal category B as a kind of generalized commuta-
tive semi-ring. Conversely, any commutative semi-ring may be viewed as a
discrete category, with only identity morphisms, which is then a symmetric
bimonoidal category.
The additive Grothendieck group completion Gr (ß0(B)) of the commu-
tative semi-ring ß0(B) is a commutative ring. Likewise, the geometric real-
ization |B| can be group completed with respect to the symmetric monoidal
pairing induced by the sum functor , and this group completion can take
place at the categorical level, say by Quillen's construction B-1B [11] or its
generalization B+ = EB xB B2 due to Thomason [31, 4.3.1]. However, the
tensor functor does not readily extend to B-1B, as was pointed out by
Thomason [32]. So B-1B is a symmetric monoidal category, but usually not
a symmetric bimonoidal category.
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 9
Example 3.1. Let V be the topological bipermutative category of finite
dimensional complex vector spaces, with set of objects N = {0, 1, 2, . .}.
with d 2 N interpreted as the complex vector space Cd, and morphism
spaces (
U(d) if d = e,
V(d, e) =
; otherwise
from d to e. The sum functor takes (d, e) to d+e and embeds U(d)xU(e)
into U(d+e) by the block sum of matrices. The tensor functor takes (d, e)
to de and maps U(d) x U(e) to U(de) by means of the left lexicographic or-
dering, which identifies {1, . .,.d}x{1, . .,.e} with {1, . .,.de}. Both of the*
*se
functors are continuous. The zero and unit objects are 0 and 1, respectively.
In this case, the semi-ring ß0(V) = N is that of the non-negative integers,
with`additive group completion Gr(N) = Z. The geometric realization |V| =
d 0BU(d) is the classifying space for complex vector bundles, while its
group completion |V-1V| ' Z x BU classifies virtual vector bundles. The
latter space is the infinite loop space underlying the spectrum ku = Spt(V)
that represents connective complex topological K-theory, which is associated
to either of the symmetric monoidal categories V or V-1V by the procedure
of Segal [27], as generalized by Shimada and Shimakawa [29] and Thomason
[31, 4.2.1].
Definition 3.2. Let (B, , , 0_, 1_) be a symmetric bimonoidal category.
The category Mn(B) of n x n matrices over B has objects the matrices
V = (Vij)ni,j=1with entries that are objects of B, and morphisms the matrices
OE = (OEij)ni,j=1with entries that are morphisms in B. The source (domain)
of OE is the matrix of sources of the entries OEij, and similarly for targets
(codomains).
There is a matrix multiplication functor
Mn(B) x Mn(B) ---.-! Mn(B)
that takes two matrices U = (Uij)ni,j=1and V = (Vjk)nj,k=1to the matrix
W = U . V = (Wik)ni,k=1with
Mn
Wik= Uij Vjk
j=1
for i, k = 1, . .,.n. In general, we need to make a definite choice of how the
n-fold sum is to be evaluated, say by bracketing from the left. When the
direct sum functor is strictly associative, as in the bipermutative case, the
choice does not matter.
The unit object In of Mn(B) is the n x n matrix with unit entries 1_on
the diagonal and zero entries 0_everywhere else.
Proposition 3.3. (Mn(B), ., In) is a monoidal category.
10 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
In other words, the functor . is associative up to a natural associativity
isomorphism
~=
ff_:U . (V . W-)---! (U . V ) . W
and unital with respect to In up to natural left and right unitality isomor-
phisms. These are coherent, in the sense that they fulfill a list of compat-
ibility conditions, including the Mac Lane-Stasheff pentagon axiom. The
proof of the proposition is a direct application of Laplaza's first coherence
theorem from [17, x7].
Definition 3.4. Let B be a commutative semi-ring with additive Grothen-
dieck group completion the commutative ring A = Gr(B). Let Mn(A) and
Mn(B) be the multiplicative monoids of n x n matrices with entries in A
and B, respectively, and let GLn(A) Mn(A) be the subgroup of invertible
n x n matrices with entries in A, i.e., those whose determinant is a unit in
A. Let the submonoid GLn(B) Mn(B) be the pullback in the diagram
GLn(B) _____//GLn(A)
fflffl fflffl
| |
fflffl| fflffl|
Mn(B) _____//_Mn(A) .
Example 3.5. When B = N and A = Z, GLn(N) = Mn(N) \ GLn(Z)
is the monoid of n x n matrices with non-negative integer entries that are
invertible as integer matrices, i.e., have determinant 1. It contains the
elementary matrices that have entries 1 on the diagonal and in one other
place, and 0 entries elsewhere. This is a larger monoid than the subgroup
of units in Mn(N), which only consists of the permutation matrices.
Definition 3.6. Let B be a symmetric bimonoidal category. Let GLn(B)
Mn(B) be the full subcategory with objects the matrices V = (Vij)ni,j=1
whose matrix of path components [V ] = ([Vij])ni,j=1lies in the submonoid
GLn(ß0(B)) Mn(ß0(B)). We call GLn(B) the category of weakly invertible
n x n matrices over B.
Corollary 3.7. (GLn(B), ., In) is a monoidal category.
Definition 3.8. Let (M, ., e) be a monoidal category, and write [p] = {0 <
1 < . .<.p}. The bar construction BM is a simplicial category [p] 7! BpM.
In simplicial degree p the category BpM has objects consisting of
(1) triangular arrays of objects Mfffiof M, for all ff < fi in [p], and
(2) isomorphisms
~=
~fffifl:Mfffi. Mfifl----!Mfffl
in M, for all ff < fi < fl in [p], such that
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 11
(3) the diagram of isomorphisms
ff_
Mfffi. (Mfifl. Mflffi)___________//(Mfffi. Mfifl) . Mflffi
id.~fiflffi|| |~fffifl.id|
fflffl| fflffl|
Mfffi. Mfiffi~fffiffi//_Mffffi~ffMfffl.fMflffilffioo_
commutes, for all ff < fi < fl < ffi in [p].
Here ff_is the associativity isomorphism for the monoidal operation . in M.
The morphisms in BpM from one object (Mfffi0, ~fffifl0) to another (Mfffi1, ~*
*fffifl1)
consist of a triangular array of morphisms OEfffi:Mfffi0! Mfffi1in M for all
ff < fi in [p], such that the diagram
~fffifl0fffl
Mfffi0. Mfifl0__//M0
ffifffi.ffififl|| ffifffl||
fflffl| fflffl|
Mfffi1. Mfifl1fffifl//_Mfffl1
~1
commutes, for all ff < fi < fl in [p].
To allow for degeneracy operators f in the following paragraph, let Mffff=
e be the unit object of M, let ~fffffiand ~fffifibe the left and right unitality
isomorphisms for ., respectively, and let OEffffbe the identity morphism on e.
The simplicial structure on BM is given as follows. For each order-
preserving function f :[q] ! [p] let the functor f* :BpM ! BqM take
the object (Mfffi, ~fffifl) of BpM to the object of BqM that consists of the
triangular array of objects Mf(ff)f(fi)for ff < fi in [q] and the isomorphisms
~f(ff)f(fi)f(fl)for ff < fi < fl in [q].
Each monoidal category M can be rigidified to an equivalent strict monoidal
category Ms, i.e., one for which the associativity isomorphism and the left
and right unitality isomorphisms are all identity morphisms [18, XI.3.1]. The
usual strict bar construction for Ms is a simplicial category [p] 7! Mps, and
corresponds in simplicial degree p to the full subcategory of BpMs where all
the isomorphisms ~fffiflare identity morphisms.
Proposition 3.9. The bar construction BM is equivalent to the strict bar
construction [p] 7! Mpsfor any strictly monoidal rigidification Ms of M.
This justifies calling BM the bar construction. The proof is an application
of Quillen's theorem A [22] and the coherence theory for monoidal categories.
Definition 3.10. Let Ar M = Fun([1], M) be the arrow category of M, with
the morphisms of M as objects and commutative square diagrams in M as
12 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
morphisms. There are obvious source and target functors s, t: Ar M ! M.
Let IsoM Ar M be the full subcategory with objects the isomorphisms of
M.
Lemma 3.11. Let (M, ., e) be a monoidal category. The category B2M is
the limit of the diagram
M x M ---.-! M ---s- IsoM ---t-! M .
For p 2 each object or morphism of BpM is uniquely determined by the
collection of its 2-faces in B2M, which is indexed by the set of monomor-
phisms f :[2] ! [p].
Consider the symmetric bimonoidal category B as a kind of generalized
semi-ring. The sum and tensor operations in B make the product category
Bn a generalized (right) module over B, for each non-negative integer n.
The collection of B-module homomorphisms Bn ! Bn is encoded in terms
of (left) matrix multiplication by the monoidal category Mn(B), and we
shall interpret the monoidal subcategory GLn(B) as a category of weak
equivalences Bn -~! Bn. This motivates the following definition.
Definition 3.12. Let B be a symmetric bimonoidal category. The algebraic
K-theory of the 2-category of (finitely generated free) modules over B is the
loop space
a
K(B) = B |BGLn(B)| .
n 0
Here |BGLn(B)| is the geometric realization of the bar construction on the
monoidal category GLn(B) of weakly invertible n x n matrices over B. The
block`sum of matrices GLn(B) x GLm (B) ! GLn+m (B) makes the coprod-
uct n 0 |BGLn(B)| a topological monoid. The looped bar construction
B provides a group completion of this topological monoid.
When B = V is the category of finite dimensional complex vector spaces,
the (finitely generated free) modules over V are called 2-vector spaces, and
K(V) is the algebraic K-theory of the 2-category of 2-vector spaces.
Let GL1 (B) = colimnGLn(B) be the infinite stabilization with respect
to block sum with the unit object in GL1(B), and write B = ß0(B) and
A = Gr (B). Then K(B) ' Z x |BGL1 (B)|+ by the McDuff-Segal group
completion theorem [21]. Here the superscript `+' refers to Quillen's plus-
construction with respect to the (maximal perfect) commutator subgroup of
GL1 (A) ~=ß1|BGL1 (B)|; cf. proposition 5.3 below.
4. Represented two-vector bundles
Let X be a topological space, with an ordered open cover (U, I). Recall
that all morphisms in V are isomorphisms, so Ar GLn(V) = IsoGLn(V).
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 13
Definition 4.1. A represented 2-vector bundle E of rank n over X consists
of
(1) a gluing map
gfffi:Ufffi----!|GLn(V)|
for each pair ff < fi in I, and
(2) a coherence map
hfffifl:Ufffifl----!|Ar GLn(V)|
satisfying s O hfffifl= gfffi. gfifland t O hfffifl= gffflover Ufffifl, *
*for each
triple ff < fi < fl in I, such that
(3) the 2-cocycle condition
hffflffiO (hfffifl. id) O ff_= hfffiffiO (id . hfiflffi)
holds over Ufffiflffifor all ff < fi < fl < ffi in I.
There is a suitably defined notion of equivalence of represented 2-vector
bundles, which we omit to formulate here, but cf. definitions 2.3 and 2.4.
Definition 4.2. Let E(d) = EU(d) xU(d)Cd # BU(d) be the universal
Cd-bundle over BU(d). There is a universal n x n matrix
E = (Eij)ni,j=1
of Hermitian vector bundles over |GLn(V)|. Over the path component
Yn
|GLn(V)D | = BU(dij)
i,j=1
for D = (dij)ni,j=1in GLn(N), the (i, j)-th entry in E is the pullback of the
universal bundle E(dij) along the projection |GLn(V)D | ! BU(dij).
Let |Ar U(d)| be the geometric realization of the arrow category Ar U(d),
where U(d) is viewed as a topological groupoid with one object. Each pair
(A, B) 2 U(d)2 defines a morphism from C to (A, B) . C = BCA-1, so
|Ar U(d)| ~=EU(d)2 xU(d)2U(d)
equals the Borel construction for this (left) action of U(d)2 on U(d). There
are source and target maps s, t: |Ar U(d)| ! BU(d), which take the 1-
simplex represented by a morphism (A, B) to the 1-simplices represented by
the morphisms A and B, respectively. By considering each element in U(d)
as a unitary isomorphism Cd ! Cd one obtains a universal unitary vector
~=
bundle isomorphism OE(d): s*E(d) -! t*E(d)
There is a universal n x n matrix of unitary vector bundle isomorphisms
OE: s*E ~=t*E
14 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
Q n
over |Ar GLn(V)|. Over the path component |Ar GLn(V)D | = i,j=1|Ar U(dij)|
for D as above, the (i, j)-th entry in OE is the pullback of the universal iso-
morphism OE(dij) along the projection |Ar GLn(V)D | ! |Ar U(dij)|.
Lemma 4.3. Let E be a represented 2-vector bundle with gluing maps gfffi
and coherence maps hfffifl. There is an associated charted 2-vector bundle
with gluing bundles
Efffi= (gfffi)*(E)
over Ufffiand coherence isomorphisms
~=
OEfffifl= (hfffifl)*(OE): Efffi. Efifl= (gfffi.-gfifl)*(E)---!(gfffl)*(E) = Ef*
*ffl
over Ufffifl. The association induces a bijection between the equivalence
classes of represented 2-vector bundles and the equivalence classes of charted
2-vector bundles of rank n over X.
Definition 4.4. Let 2-Vectn(X) be the set of equivalence classes of 2-vector
bundles of rank n over X. For path-connected X let
a
2-Vect(X) = 2-Vectn(X) .
n 0
Whitney sum (definition 2.7) defines a pairing that makes 2-Vect(X) an
abelian monoid.
Theorem 4.5. Let X be a finite CW complex. There are natural bijections
2-Vectn(X) ~=[X, |BGLn(V)|]
and a
2-Vect(X) ~=[X, |BGLn(V)|] .
n 0
To explain the first correspondence, from which the second follows, we
use the following construction.
Definition 4.6. Let (U, I) be an ordered open cover of X. The Mayer-
Vietoris blow-up MV (U) of X with respect to U is the simplicial space with
p-simplices a
MVp(U) = Uff0...ffp
ff0 ... ffp
with ff0 . . . ffp in I. The i-th simplicial face map is a coproduct
of inclusions Uff0...ffp Uff0...^ffi...ffp, and similarly for the degeneracy m*
*aps.
The inclusions Uff0...ffp X combine to a natural map e: |MV (U)| ! X,
which is a (weak) homotopy equivalence.
Sketch of proof of theorem 4.5. By lemma 3.11, a simplicial map g :MV (U) !
|BGLn(V)| is uniquely determined by its components in simplicial degrees
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 15
1 and 2. The first of these is a map
`
g1: MV1(U) = ff fiUfffi----! |B1GLn(V)| = |GLn(V)|
which is a coproduct of gluing maps gfffi:Ufffi! |GLn(V)|. The second is a
map
a
g2: MV2(U) = Ufffifl! |B2GLn(V)| .
ff fi fl
The simplicial identities and lemma 3.11 imply that g2 is determined by g1
and a coproduct of coherence maps hfffifl:Ufffifl! |Ar GLn(V)|. Hence such
a simplicial map g corresponds bijectively to a represented 2-vector bundle
of rank n over X.
Any map f :X ! |BGLn(V)| can be composed with the weak equivalence
e: |MV (U)| ! X to give a map of spaces fe: |MV (U)| ! |BGLn(V)|, which
is homotopic to a simplicial map g if U is a good cover, and for reasonable
X any open ordered cover can be refined to a good one. The homotopy class
of f corresponds to the equivalence class of the represented 2-vector bundle
determined by the simplicial map g.
Remark 4.7. We wish to interpret the 2-vector bundles over X as (effective)
0-cycles for some cohomology theory at X. Such theories are group-valued,
so a first approximation to the 0-th cohomology group at X could be the
Grothendieck group Gr(2-Vect(X)) of formal differences of 2-vector bundles
over X. The analogous construction for ordinary vector bundles works well
to define topological K-theory, but for 2-vector bundles this algebraically
group completed functor is not even representable, like in the case of the
algebraic K-theory of a discrete ring. We thank Haynes Miller for reminding
us of this issue.
Instead we follow Quillen and perform the group completion at the space
level, which leads to the algebraic K-theory space
a
K(V) = B |BGLn(V)|
n 0
' Z x |BGL1 (V)|+
from definition 3.12. But what theory does this loop space represent? One
interpretation is provided by the theory of virtual flat fibrations, presented
by Karoubi in [16, Ch. III], leading to what we shall call virtual 2-vector
bundles. Another interpretation could be given using the homology bordism
theory of Hausmann and Vogel [12].
Definition 4.8. Let X be a space. An acyclic fibration over X is a Serre
fibration a: Y ! X such that the homotopy fiber at each point x 2 X has
the integral homology of a point, i.e., ~H*(hofibx(a); Z) = 0. A map of acyclic
fibrations from a0:Y 0! X to a: Y ! X is a map f :Y 0! Y with af = a0.
16 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
A virtual 2-vector bundle over X is described by an acyclic fibration
a: Y ! X and a 2-vector bundle E # Y . We write E # Y -a! X. Given
a map f :Y 0! Y of acyclic fibrations over X there is an induced 2-vector
bundle f*E # Y 0. The virtual 2-vector bundles described by E # Y - a! X
0
and f*E # Y 0-a! X are declared to be equivalent as virtual 2-vector bundles
over X.
Lemma 4.9. The abelian monoid of equivalence classes of virtual 2-vector
bundles over X is the colimit
colim 2-Vect(Y )
a:Y !X
where a: Y ! X ranges over the category of acyclic fibrations over X. Its
Grothendieck group completion is isomorphic to the colimit
colim Gr(2-Vect(Y )) .
a:Y !X
The functor Y 7! 2-Vect(Y ) factors through the homotopy category of
acyclic fibrations over X, which is directed.
The following result says that formal differences of virtual 2-vector bun-
dles over X are the geometric objects that constitute cycles for the con-
travariant homotopy functor represented by the algebraic K-theory space
K(V). Compare [16, III.3.11]. So K(V) represents sheaf cohomology for
the topology of acyclic fibrations, with coefficients in the abelian presheaf
Y 7! Gr (2-Vect(Y )) given by the Grothendieck group completion of the
abelian monoid of equivalence classes of 2-vector bundles.
Theorem 4.10. Let X be a finite CW complex. There is a natural group
isomorphism
colim Gr (2-Vect(Y )) ~=[X, K(V)]
a:Y !X
where a: Y ! X ranges over the category of acyclic fibrations over X.
Restricted to Gr (2-Vect(X)) (with a = id) the isomorphism`extends the
canonical monoid homomorphism 2-Vect(X) ~=[X, n 0|BGLn(V)|] !
[X, K(V)].
Remark 4.11. The passage to sheaf cohomology would be unnecessary if we
replaced V by a different symmetric bimonoidal category B such that each
ß0(GLn(B)) is abelian. This might entail an extension of the category of
vector spaces to allow generalized vector spaces of arbitrary real, or even
complex, dimension, parallel to the inclusion of the integers into the real or
complex numbers. Such an extension is reminiscent of a category of repre-
sentations of a suitable C*-algebra, but we know of no clear interpretation
of this approach.
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 17
5. Algebraic K-theory of topological K-theory
Is the contravariant homotopy functor X 7! [X, K(V)] = K(V)0(X) part
of a cohomology theory, and if so, what is the spectrum representing that
theory?
The topological symmetric bimonoidal category V plays the role of a gen-
eralized commutative semi-ring in our definition of K(V). Its additive group
completion V-1V correspondingly plays the role of a generalized commuta-
tive ring. This may be tricky to realize at the level of symmetric bimonoidal
categories, but the connective topological K-theory spectrum ku = Spt(V)
associated to the additive topological symmetric monoidal structure of V
is an E1 ring spectrum, and hence a commutative algebra over the sphere
spectrum S.
The algebraic K-theory of an S-algebra A can on one hand be defined as
the Waldhausen algebraic K-theory [34] of a category with cofibrations and
weak equivalences, with objects the finite cell A-modules, morphisms the A-
module maps and weak equivalences the stable equivalences. Alternatively,
it can be defined as a group completion
a
K(A) = B BdGLn (A)
n 0
where dGLn(A) is essentially the topological monoid of A-module maps An !
An that are stable equivalences. The former definition produces a spectrum,
so the space K(A) is in fact an infinite loop space, and its deloopings rep-
resent a cohomology theory.
The passage from modules over the semi-ring object V to modules over
the ring object ku corresponds to maps |GLn(V)| ! dGLn(ku) and a map
K(V) ! K(ku).
Conjecture 5.1. There is a weak equivalence K(V) ' K(ku). More gen-
erally, K(B) ' K(A) for each symmetric bimonoidal category B with asso-
ciated commutative S-algebra A = Spt(B).
Remark 5.2. The conjecture asserts that the contravariant homotopy functor
X 7! [X, K(V)] with 0-cycles given by the virtual 2-vector bundles over X
is the 0-th cohomology group for the cohomology theory represented by the
spectrum K(ku) given by the algebraic K-theory of connective topological
K-theory. We consider the virtual 2-vector bundles over X to be sufficiently
geometric objects (like complex vector bundles), that this cohomology the-
ory then admits as geometric an interpretation as the classical examples of
de Rham cohomology, topological K-theory and complex bordism.
As a first (weak) justification of this conjecture, recall that to the eyes
of algebraic K-theory the block sum operation (g, h) 7! g00h is identified
with the stabilized matrix multiplication (g, h) 7! gh00I, where I is an
18 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
identity matrix. The group completion in the definition of algebraic K-
theory adjoins inverses to the block sum operation, and thus also to the
stabilized matrix multiplication. In particular, for each elementary n x n
matrix eij(V ) over B with (i, j)-th off-diagonal entry equal to an object V
of B, the inverse matrix eij(-V ) is formally adjoined as far as algebraic K-
theory is concerned. Hence the formal negatives -V in B-1B are already
present, in this weak sense.
A stronger indication that the conjecture should hold true is provided by
the following special case. Recall that a commutative semi-ring is the same
as a (small) symmetric bimonoidal category that is discrete, i.e., has only
identity morphisms.
Proposition 5.3. Let B be a commutative semi-ring, with additive Groth-
endieck group completion A = Gr(B). The semi-ring homomorphism B !
A induces a weak equivalence
BGL1 (B) --'--! BGL1 (A)
and thus a weak equivalence K(B) ' K(A). In particular, there is a weak
equivalence K(N) ' K(Z).
A proof uses the following application of Quillen's theorem B [22].
Lemma 5.4. Let f :M ! G be a monoid homomorphism from a monoid
M to a group G. Write mg = f(m) . g. Let Q = B(*, M, G) be the category
with objects g 2 G and morphisms (m, g) 2 M x G from mg to g:
(m,g)
mg ----! g .
Then there is a fiber sequence up to homotopy
|Q| ----! BM --Bf--!BG .
Sketch of proof of proposition 5.3. Applying lemma 5.4 to the monoids
Mn = GLn(B) and groups Gn = GLn(A) we obtain categories Qn for each
natural number n. There are stabilization maps i: Qn ! Qn+1, Mn !
Mn+1 and Gn ! Gn+1, with (homotopy) colimits Q1 , M1 and G1 , and a
quasi-fibration
|Q1 | ----! BGL1 (B) ----! BGL1 (A) .
It suffices to show that each stabilization map i: |Qn| ! |Qn+1| is weakly
null-homotopic, because then |Q1 | is weakly contractible.
For each full subcategory K Qn with finitely many objects, the re-
stricted stabilization functor i|K takes g to i(g) = g001. It receives a natu-
ral transformation from a functor j :K ! Qn+1 that maps g to j(g) = [gv01]
for some column vector v = v(g) with positive entries in B. The trick is to
construct v(g) inductively for the finite set of objects g of K, so that v(mg)
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 19
is sufficiently positive with respect to m . v(g) for all morphisms mg ! g in
K.
Furthermore, the finiteness of K ensures that there is a row vector w
with entries in B and an object h = In-0w1of Qn+1 such that there is
a natural transformation from j to the constant functor to h. These two
natural transformations provide a homotopy from i|K to a constant map.
As K was arbitrary with finitely many objects, this means that i is weakly
null-homotopic.
Remark 5.5. If there exists a symmetric bimonoidal category W and a func-
tor V ! W of symmetric bimonoidal categories that induces an additive
equivalence from V-1V to W, then most likely the line of argument sketched
above in the case of commutative semi-rings can be adapted to the sym-
metric bimonoidal situation. This could provide one line of proof toward
conjecture 5.1. Similar remarks apply for a general symmetric bimonoidal
category B in place of V.
6.Forms of elliptic cohomology
In this section we shall view the algebraic K-theory K(A) of an S-algebra
A as a spectrum, rather than as a space.
We shall argue that the algebraic K-theory K(ku) of the connective topo-
logical K-theory spectrum ku is a connective form of elliptic cohomology,
in the sense that it detects homotopy theoretic phenomena related to v2-
periodicity, much like how topological K-theory detects phenomena related
to v1-periodicity (which is really the same as Bott periodicity) and how
rational cohomology detects phenomena related to v0-periodicity. Further-
more, from this point of view the homotopy type of K(ku) is robust with
respect to changes in the interpretation of the phrase ä lgebraic K-theory
of topological K-theory".
We first introduce a filtration of the class of spectra that is related to
the chromatic filtration given by the property of being Bousfield local with
respect to some Johnson-Wilson theory E(n) (cf. Ravenel [24, x7]), but is
more appropriate for the connective spectra that arise from algebraic K-
theory. Our notion is also more closely linked to aspects of vn-periodicity
than to being E(n)-local.
Let p be a prime, K(n) the n-th Morava K-theory at p and F a p-local
finite CW spectrum. The least number 0 n < 1 such that K(n)*(F )
is non-trivial is called the chromatic type of F . (Only contractible spectra
have infinite chromatic type.) By the Hopkins-Smith periodicity theorem
[14, Thm. 9], F admits a vn-self map v : dF ! F such that K(m)*(v) is an
isomorphism for m = n and zero for m 6= n. The vn-self map is sufficiently
unique for the mapping telescope
v v
v-1 F = Tel F ----! -dF ----! . . .
20 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
to be well-defined up to homotopy. The class of all p-local finite CW spectra
of chromatic type n is closed under weak equivalences and the formation
of homotopy cofibers, desuspensions and retracts, so we say that the full
subcategory that it generates is a thick subcategory. By the Hopkins-Smith
thick subcategory theorem [14, Thm. 7], any thick subcategory of the cate-
gory of p-local finite CW spectra has this precise form, for a unique number
0 n 1.
Definition 6.1. Let X be a spectrum, and let TX be the full subcategory
of p-local finite CW spectra F for which the localization map
F ^ X ----! v-1 F ^ X
induces an isomorphism on homotopy groups in all sufficiently high degrees.
Then TX is a thick subcategory, hence consists of the spectra F of chromatic
type n for some unique number 0 n 1. We call this number n =
telecom(X) the telescopic complexity of X. (This abbreviation is due to
Matthew Ando.)
Lemma 6.2. If Y is the k-connected cover of X, for some integer k, then
X and Y have the same telescopic complexity.
Let X ! Y ! Z be a cofiber sequence. If telecom(X) 6= telecom(Y )
then telecom(Z) = max {telecom(X), telecom(Y )}, otherwise telecom(Z)
max {telecom(X), telecom(Y )}.
If Y is a (de-)suspension of X then X and Y have the same telescopic
complexity.
If Y is a retract of X then telecom(Y ) telecom(X).
If X is an E(n)-local spectrum then X has telescopic complexity n.
Examples 6.3. (1) Integral, rational, real and complex cohomology (HZ,
HQ, HR or HC) all have telescopic complexity 0.
(2) Connective or periodic, real or complex topological K-theory (ko, ku,
KO or KU) all have telescopic complexity 1. The 'etale K-theory Ket(R) of
a ring R = OF,S of S-integers in a local or global number field has telescopic
complexity 1, and so does the algebraic K-theory K(R) if the Lichtenbaum-
Quillen conjecture holds for the ring R.
(3) An Ando-Hopkins-Strickland [1] elliptic spectrum (E, C, t) has tele-
scopic complexity 2, and the telescopic complexity equals 2 if and only if
the elliptic curve C over R = ß0(E) has a supersingular specialization over
some point of Spec(R).
(4) The Hopkins-Mahowald-Miller topological modular forms spectra tmf
and TMF have telescopic complexity 2.
(5) The Johnson-Wilson spectrum E(n) and its connective form, the
Brown-Peterson spectrum BP , both have telescopic complexity n.
(6) The sphere spectrum S and the complex bordism spectrum MU have
infinite telescopic complexity.
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 21
Let V (1) be the four-cell Smith-Toda spectrum with
BP*(V (1)) = BP*=(p, v1) .
For p 5 it exists as a commutative ring spectrum. It has chromatic type 2,
and there is a v2-self map v : 2p2-2V (1) ! V (1) inducing multiplication
by the class v2 2 ß2p2-2V (1). We write V (1)*(X) = ß*(V (1) ^ X) for
the V (1)-homotopy groups of X, which are naturally a graded module over
P (v2) = Fp[v2].
Let X(p)and X^pdenote the p-localization and p-completion of a spec-
trum X, respectively. The first Brown-Peterson spectrum ` = BP <1> is the
connective p-local Adams summand of ku(p), and its p-completion `^pis the
connective p-complete Adams summand of ku^p. These are all known to be
commutative S-algebras.
The spectrum T C(`^p) occurring in the following statement is the topolog-
ical cyclic homology of `^p, as defined by Bökstedt, Hsiang and Madsen [5].
The theorem is proved in [3, 0.3] by an elaborate but explicit computation of
its V (1)-homotopy groups, starting from the corresponding V (1)-homotopy
groups of the topological Hochschild homology T HH(`^p).
Theorem 6.4 (Ausoni-Rognes). Let p 5. The algebraic K-theory
spectrum K(`^p) of the connective p-complete Adams summand `^phas tele-
scopic complexity 2. More precisely, there is an exact sequence of P (v2)-
modules
0 ! 2p-3Fp ! V (1)*K(`^p) --trc--!V (1)*T C(`^p) ! -1Fp ! 0
and an isomorphism of P (v2)-modules
V (1)*T C(`^p) ~=P (v2)
d dp
E(@, ~1, ~2) E(~2){~1t | 0 < d < p} E(~1){~2t | 0 < d < p} .
Here @, t, ~1 and ~2 have degrees -1, -2, 2p - 1 and 2p2 - 1, respectively.
Hence V (1)*T C(`^p) is free of rank (4p + 4) over P (v2), and agrees with its
v2-localization in sufficiently high degrees.
Since K(`^p) has telescopic complexity 2, it has a chance to detect v2-
periodic families in ß*V (1). This is indeed the case. Let ff1 2 ß2p-3V (1)
and fi012 ß2p2-2p-1V (1) be the classes represented in the Adams spectral
sequence by the cobar 1-cycles h10 = [~,1] and h11 = [~,p1], respectively. There
are maps V (1) ! v-12V (1) ! L2V (1), and Ravenel [23, 6.3.22] computed
ß*L2V (1) ~=P (v2, v-12) E(i){1, h10, h11, g0, g1, h11g0 = h10g1}
for p 5. Hence ß*L2V (1) contains twelve v2-periodic families. The tele-
scope conjecture asserted that v-12V (1) ! L2V (1) might be an equivalence,
but this is now considered to be unlikely [19]. The following detection result
22 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
can be read off from [3, 4.8], and shows that K(`^p) detects the same kind
of homotopy theoretic phenomena as E(2) or an elliptic spectrum.
Proposition 6.5. The unit map S ! K(`^p) induces a P (v2)-module ho-
momorphism ß*V (1) ! V (1)*K(`^p) which takes 1, ff1 and fi01to 1, t~1
and tp~2, respectively. Hence V (1)*K(`^p) detects the v2-periodic families
in ß*V (1) generated by these three classes.
Turning to the whole connective p-complete topological K-theory spec-
trum ku^p, there is a map `^p! ku^pof commutative S-algebras. It induces a
natural map K(`^p) ! K(ku^p), and there is a transfer map K(ku^p) ! K(`^p)
such that the composite self-map of K(`^p) is multiplication by (p-1). Hence
the composite map is a p-local equivalence.
Lemma 6.6. The algebraic K-theory spectrum K(ku^p) of connective p-
complete topological K-theory ku^pcontains K(`^p) as a p-local retract, hence
has telescopic complexity 2.
Most likely K(ku^p) also has telescopic complexity exactly 2. It may
be possible to prove this directly by computing V (1)*T C(ku^p), by similar
methods as in [3], but the algebra involved for ku^pis much more intricate
than it was for the Adams summand. Some progress in this direction has
recently been made by Ausoni.
The following consequence of a theorem of the second author [9, p. 224]
allows us to compare the algebraic K-theory of ku^pto that of the integral
spectra ku and K(C).
Theorem 6.7 (Dundas). Let A be a connective S-algebra. The commu-
tative square
K(A) ________//_K(A^p)
| |
| |
fflffl| fflffl|
K(ß0(A)) ____//_K(ß0(A^p))
becomes homotopy Cartesian after p-completion.
We apply this with A = ku or A = K(C). Also in the second case
A^p' ku^p, by Suslin's theorem on the algebraic K-theory of algebraically
closed fields [30]. Then ß0(A) = Z and ß0(A^p) = Zp. It is known that K(Zp)
has telescopic complexity 1, by Bökstedt-Madsen [6], [7] for p odd and by
the third author [26] for p = 2. It is also known that K(Z) has telescopic
complexity 1 for p = 2, by Voevodsky's proof of the Milnor conjecture
and Rognes-Weibel [25]. For p odd it would follow from the Lichtenbaum-
Quillen conjecture for K(Z) at p that K(Z) has telescopic complexity 1, and
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 23
this now seems to be close to a theorem by the work of Voevodsky, Rost and
Positselski.
Proposition 6.8. Suppose that K(Z) has telescopic complexity 1 at a prime
p 5. Then K(ku) and K(K(C)) have the same telescopic complexity as
K(ku^p), which is 2.
More generally it is natural to expect that K(K(R)) has telescopic com-
plexity 2 for each ring of S-integers R = OF,S in a local or global number
field F , including the initial case K(K(Z)). A discussion of such a conjec-
ture has been given in lectures by the third author, but should take place in
the context of 'etale covers or Galois extensions of commutative S-algebras,
which would take us too far afield here.
The difference between the connective and periodic topological K-theory
spectra ku and KU may also not affect their algebraic K-theories greatly.
There is a (localization) fiber sequence
K(CKU (ku)) ! K(ku) ! K(KU)
where CKU (ku) is the category of finite cell ku-module spectra that become
contractible when induced up to KU-modules [34, 1.6.4]. Such spectra have
finite Postnikov towers with layers that are induced from finite cell HZ-
module spectra via the map ku ! HZ, and so it is reasonable to expect that
a generalized form of the devissage theorem in algebraic K-theory applies
to identify K(CKU (ku)) with K(Z).
Proposition 6.9. If there is a fiber sequence K(Z) ! K(ku) ! K(KU)
and K(Z) has telescopic complexity 1, at a prime p 5, then the algebraic
K-theory spectrum K(KU) of the periodic topological K-theory spectrum
KU has the same telescopic complexity as K(ku), which is 2.
Remark 6.10. Unlike traditional elliptic cohomology, the spectrum K(ku)
is not complex orientable. For the unit map of K(Z) detects j 2 ß1(S)
and factors as S ! K(ku) ! K(Z), where the first map is the unit of
K(ku) and the second map is induced by the map ku ! HZ of S-algebras.
Hence the unit map for K(ku) detects j and cannot factor through the
complex bordism spectrum MU, since ß1(MU) = 0. This should not be
perceived as a problem, however, as e.g. also the topological modular_forms_
spectrum tmf is not complex orientable._It seems more likely that K(KU )
can be complex oriented, where KU is an ä lgebraic closureö f KU in the
category of commutative S-algebras.
References
[1]M. Ando, M. J. Hopkins, and N. P. Strickland. Elliptic spectra, the Witten *
*genus
and the theorem of the cube. Invent. Math., 146(3):595-687, 2001.
24 NILS A. BAAS, BJfflRN IAN DUNDAS AND JOHN ROGNES
[2]M. F. Atiyah and G. B. Segal. The index of elliptic operators. II. Ann. of *
*Math. (2),
87:531-545, 1968.
[3]Christian Ausoni and John Rognes. Algebraic K-theory of topological K-theor*
*y. Acta
Math., 188(1):1-39, 2002.
[4]Nils A. Baas. A general topological functor construction. 1998. NTNU prepri*
*nt,
Trondheim.
[5]M. Bökstedt, W. C. Hsiang, and I. Madsen. The cyclotomic trace and algebrai*
*c K-
theory of spaces. Invent. Math., 111(3):465-539, 1993.
[6]M. Bökstedt and I. Madsen. Topological cyclic homology of the integers. Ast*
*'erisque,
(226):7-8, 57-143, 1994. K-theory (Strasbourg, 1992).
[7]M. Bökstedt and I. Madsen. Algebraic K-theory of local number fields: the u*
*nramified
case. In Prospects in topology (Princeton, NJ, 1994), volume 138 of Ann. of *
*Math.
Stud., pages 28-57. Princeton Univ. Press, Princeton, NJ, 1995.
[8]Jean-Luc Brylinski. Loop spaces, characteristic classes and geometric quant*
*ization,
volume 107 of Progress in Mathematics. Birkhäuser Boston Inc., Boston, MA, 1*
*993.
[9]Bjørn Ian Dundas. Relative K-theory and topological cyclic homology. Acta M*
*ath.,
179(2):223-242, 1997.
[10]Jean Giraud. Cohomologie non ab'elienne. Springer-Verlag, Berlin, 1971. Die*
* Grund-
lehren der mathematischen Wissenschaften, Band 179.
[11]Daniel Grayson. Higher algebraic K-theory. II (after Daniel Quillen). In Al*
*gebraic
K-theory (Proc. Conf., Northwestern Univ., Evanston, Ill., 1976), pages 217-*
*240.
Lecture Notes in Math., Vol. 551. Springer, Berlin, 1976.
[12]Jean-Claude Hausmann and Pierre Vogel. The plus construction and lifting ma*
*ps from
manifolds. In Algebraic and geometric topology (Proc. Sympos. Pure Math., St*
*anford
Univ., Stanford, Calif., 1976), Part 1, Proc. Sympos. Pure Math., XXXII, pag*
*es
67-76. Amer. Math. Soc., Providence, R.I., 1978.
[13]Nigel Hitchin. Lectures on special Lagrangian submanifolds. In Winter Schoo*
*l on
Mirror Symmetry, Vector Bundles and Lagrangian Submanifolds (Cambridge, MA,
1999), volume 23 of AMS/IP Stud. Adv. Math., pages 151-182. Amer. Math. Soc.,
Providence, RI, 2001.
[14]Michael J. Hopkins and Jeffrey H. Smith. Nilpotence and stable homotopy the*
*ory. II.
Ann. of Math. (2), 148(1):1-49, 1998.
[15]M. M. Kapranov and V. A. Voevodsky. 2-categories and Zamolodchikov tetrahe-
dra equations. In Algebraic groups and their generalizations: quantum and in*
*finite-
dimensional methods (University Park, PA, 1991), volume 56 of Proc. Sympos. *
*Pure
Math., pages 177-259. Amer. Math. Soc., Providence, RI, 1994.
[16]Max Karoubi. Homologie cyclique et K-th'eorie. Ast'erisque, (149):147, 1987.
[17]Miguel L. Laplaza. Coherence for distributivity. In Coherence in categories*
*, pages
29-65. Lecture Notes in Math., Vol. 281. Springer, Berlin, 1972.
[18]Saunders MacLane. Categories for the working mathematician. Springer-Verlag*
*, New
York, 1971. Graduate Texts in Mathematics, Vol. 5.
[19]Mark Mahowald, Douglas Ravenel, and Paul Shick. The triple loop space appro*
*ach to
the telescope conjecture. In Homotopy methods in algebraic topology (Boulder*
*, CO,
1999), volume 271 of Contemp. Math., pages 217-284. Amer. Math. Soc., Provid*
*ence,
RI, 2001.
[20]J. Peter May. E1 ring spaces and E1 ring spectra. Springer-Verlag, Berlin*
*, 1977.
With contributions by Frank Quinn, Nigel Ray, and Jørgen Tornehave, Lecture *
*Notes
in Mathematics, Vol. 577.
[21]D. McDuff and G. Segal. Homology fibrations and the "group-completion" theo*
*rem.
Invent. Math., 31(3):279-284, 1975/76.
[22]Daniel Quillen. Higher algebraic K-theory. I. In Algebraic K-theory, I: Hig*
*her K-
theories (Proc. Conf., Battelle Memorial Inst., Seattle, Wash., 1972), pages*
* 85-147.
Lecture Notes in Math., Vol. 341. Springer, Berlin, 1973.
TWO-VECTOR BUNDLES AND FORMS OF ELLIPTIC COHOMOLOGY 25
[23]Douglas C. Ravenel. Complex cobordism and stable homotopy groups of spheres*
*, vol-
ume 121 of Pure and Applied Mathematics. Academic Press Inc., Orlando, FL, 1*
*986.
[24]Douglas C. Ravenel. Nilpotence and periodicity in stable homotopy theory, v*
*olume 128
of Annals of Mathematics Studies. Princeton University Press, Princeton, NJ,*
* 1992.
Appendix C by Jeff Smith.
[25]J. Rognes and C. Weibel. Two-primary algebraic K-theory of rings of integer*
*s in
number fields. J. Amer. Math. Soc., 13(1):1-54, 2000. Appendix A by Manfred *
*Kol-
ster.
[26]John Rognes. Algebraic K-theory of the two-adic integers. J. Pure Appl. Alg*
*ebra,
134(3):287-326, 1999.
[27]Graeme Segal. Categories and cohomology theories. Topology, 13:293-312, 197*
*4.
[28]Graeme Segal. Elliptic cohomology (after Landweber-Stong, Ochanine, Witten,*
* and
others). Ast'erisque, (161-162):Exp. No. 695, 4, 187-201 (1989), 1988. S'emi*
*naire Bour-
baki, Vol. 1987/88.
[29]Nobuo Shimada and Kazuhisa Shimakawa. Delooping symmetric monoidal categori*
*es.
Hiroshima Math. J., 9(3):627-645, 1979.
[30]A. Suslin. On the K-theory of algebraically closed fields. Invent. Math., 7*
*3(2):241-
245, 1983.
[31]R. W. Thomason. Homotopy colimits in the category of small categories. Math*
*. Proc.
Cambridge Philos. Soc., 85(1):91-109, 1979.
[32]R. W. Thomason. Beware the phony multiplication on Quillen's A-1A. Proc. Am*
*er.
Math. Soc., 80(4):569-573, 1980.
[33]Friedhelm Waldhausen. Algebraic K-theory of topological spaces. I. In Algeb*
*raic
and geometric topology (Proc. Sympos. Pure Math., Stanford Univ., Stanford, *
*Calif.,
1976), Part 1, Proc. Sympos. Pure Math., XXXII, pages 35-60. Amer. Math. Soc*
*.,
Providence, R.I., 1978.
[34]Friedhelm Waldhausen. Algebraic K-theory of spaces. In Algebraic and geomet*
*ric
topology (New Brunswick, N.J., 1983), volume 1126 of Lecture Notes in Math.,*
* pages
318-419. Springer, Berlin, 1985.
Department of Mathematical Sciences, NTNU, Trondheim, Norway
E-mail address: baas@math.ntnu.no
Department of Mathematical Sciences, NTNU, Trondheim, Norway
E-mail address: dundas@math.ntnu.no
Department of Mathematics, University of Oslo, Norway
E-mail address: rognes@math.uio.no